Coverage Noise

[mbillingr]

Coverage increases and decreases between builds by +/- 13 lines in _infomax.py. I've tracked this down to the if wts_blowup: code branch. Sometimes the weights blow up and we get these 13 lines, and somitemes they don't and we miss those lines.

Maybe we did not random-seed all the tests yet? Or does Infomax ignore our seeds?

Also, I noticed that all code branches related to Extended Infomax are untested. Haven't we changed the default to use extended, or did we just discuss this?

[cbrnr]

We only discussed changing to Extended Infomax in #53. There were still a couple of open questions.

Infomax does have a seed argument, so it shouldn't ignore it if we set it. It's more likely we've not seeded all examples.

[mbillingr]

I have never seeded Infomax in particular. The tests only set the seed of the global numpy random engine.

At least one test does not set the seed: TestMVARICA.testModelIdentification() in test_varica.py. There may be more, still.